Research On Learning-based Image Super Resolution Reconstruction Method

Due to the hardware limitations of the imaging device and the natural factors in the imaging process,naturally acquired images tend to exhibit lower resolution.Improving the resolution of the image through hardware requires increasing the size of the sensor chip and the number of pixels in per unit area.This requires not only high costs,but also technical challenges.Super-resolution image reconstruction is a technology to eliminate the image degradation in the imaging process through signal processing methods.It can overcome the limitations of hardware conditions and improve image resolution by using the approach of software.Nowadays,with the rapid development of information technology,super-resolution image reconstruction technology has important applications in many fields and has become a research hot in computer vision.There are also special applications in military satellites,public security and transportation,video surveillance and forensics,medical imaging,industrial imaging,and so on.In this paper,we focus on learning-based super-resolution image reconstruction methods,including super-resolution image reconstruction techniques based on traditional machine learning and super-resolution image reconstruction methods based on deep learning.The main works are as follows:A super-resolution image reconstruction method based on multi-feature fusion and sparse representation is proposed.In the proposed method,we use sparse dictionary to reconstruct the details of the luminance channel and restore the high-frequency components for the color image,and use edge interpolation method to reconstruct the CbCr channel of the image,which can effectively improve the definition of the image edge.To reconstruct the luminance channel,we propose a joint feature representation method that combines multiple features to reflect the high-frequency information of the image from multiple aspects.Moreover,according to the visual characteristics that the human eye is more sensitive to high-frequency information,the over-smooth regions in image are removed,and the salient regions with high-frequency information are retained,and the final training sample can be obtained.Through joint training and learning,the sparse representation of the high resolution image block is the same as the sparse representation of the corresponding low resolution image block.In the process of image reconstruction,we combine the local weighted constraint regularization term to improve the speed and precision of solving sparse matrix sparse solution,and final,the reconstructed high resolution image can be obtained.The experimental results show that the proposed method has a satisfactory reconstruction result,and the visual effects and numerical evaluation results of the reconstructed images are more superior to the comparative methods.With the successful application of the deep convolutional neural networks in computer vision fields such as image target recognition and image classification.In 2015,Dong introduced the convolutional neural network into the research of image super-resolution reconstruction,and achieved good research results.In order to improve the quality of the reconstructed image,we propose an improved method based on Dong's algorithm.Our improvement is mainly reflected in two aspects:the first is to improve the network structure.By increasing the depth of the network and the number of convolutional layers,the network model can learn more comprehensive information so as to obtain more abundant and more representative image features.The second is to present a new nonlinear activation function that combine with the rectified linear unit activation function and softplus activation function.The new nonlinear activation function is more similar to the activation characteristics of biological neurons.At the same time,it optimizes the network performance by making the neural network more sparse.The experimental results show that,compared with the original method,the improved method has a better reconstruction effect.